以衛星資料預測天災及期貨價格變化;Predicting natural disaster and commodity price movement with satellite data

NCUIR > School of Management at National Central University > Graduate Institute of Business Administration > Electronic Thesis & Dissertation > Item 987654321/92428

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/92428

Title:	以衛星資料預測天災及期貨價格變化;Predicting natural disaster and commodity price movement with satellite data
Authors:	曾柔慈;Tseng, Jou-Tzu
Contributors:	企業管理學系
Keywords:	衛星遙測;植被指數;區域性異常因子;自編碼器;Satellite telemetry;Normalized Difference Vegetation Index;Local Outlier Factor;Auto-Encoder
Date:	2024-01-30
Issue Date:	2024-09-19 15:51:34 (UTC+8)
Publisher:	國立中央大學
Abstract:	本研究欲通過結合遙感技術和機器學習的方法，旨在探討天災對美國主要玉米產區的歸一化植被指數 (NDVI) 變化，以及其對於美國玉米期貨市場價格的潛在影響。為此，本研究提出一種基於結合自編碼器 (Auto-Encoder) 和區域異常因子 (Local Outlier Factor, LOF) 的模型，先利用Auto-Encoder進行特徵學習，捕捉數據中關鍵的特徵值，接著使用訓練完成的Auto-Encoder中Encoder的部分將原始數據集進行轉換，作為LOF模型的輸入、並訓練模型進行異常值檢測 (天災預測) ，最後通過多階段的模型參數調整，尋求最佳的參數配置和異常值閾值設定。實驗結果表明，我們所提出的模型在天災預測方面達到了65% 的Precision、69%的Recall、以及67% 的F1 Score。除此之外，在模型準確預測的天災案例中，我們觀察到天災發生當日對美國玉米期貨市場價格的影響，無論是從期貨交易筆數、還是從整體價格的漲/跌幅的角度來看，我們發現價格多數呈現上漲趨勢。綜合上述，本研究展示了結合遙測技術和機器學習在農業監測和災害管理領域的應用潛力。;Our study aims to explore the impact of natural disasters on the Normalized Difference Vegetation Index (NDVI) in major corn-producing areas of the United States, and its potential influence on the US corn futures market, through the integration of remote sensing technology and machine learning methods. To this end, we propose a model that combines an Auto-Encoder and Local Outlier Factor (LOF). Initially, the Auto-Encoder is utilized for feature learning to capture key characteristics within the data. Then, the trained Auto-Encoder′s encoder is used to transform the original dataset, serving as the input for the LOF model for anomaly detection (predicting natural disasters). Finally, through multi-stage parameter adjustments, our study seeks the optimal configuration of parameters and anomaly threshold settings. The experimental results indicate that our proposed model achieved 65% Precision, 69% Recall, and 67% F1 Score in disaster prediction. Furthermore, in the disaster cases accurately predicted by the model, we observed the impact on the United States corn futures market price on the day of the disaster. From the perspective of both futures trading volume and overall price fluctuations (rising or falling), we found that prices generally showed an upward trend. In summary, our study demonstrates the potential application of combining remote sensing technology and machine learning in the fields of agricultural monitoring and disaster management.
Appears in Collections:	[Graduate Institute of Business Administration] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	64	View/Open

社群 sharing

Loading...